Real-time ensemble data assimilation at the University of Washington

One method for quantifying the error in a weather forecast involves integrating an ensemble of forecasts forward in time. However, since operational centers currently employ techniques that only produce deterministic analyses, perturbation techniques are needed to populate an ensemble of initial conditions for use in ensemble forecasting. Recently, the ensemble Kalman filter (EnKF) has emerged as a possible alternative to these practices because it produces an ensemble of equally likely analyses during observation assimilation that may then be used for ensemble forecasting. Moreover, since the ensemble contains probabilistic information about the state, observations are assimilated in a flow-dependent manner, unlike the fixed statistics used in current operational methods.

Since 23 December 2004, a 90-member real-time EnKF has been running at the University of Washington using the Weather Research and Forecasting (WRF) model over the eastern Pacific and western North America (http://www.atmos.washington.edu/~enkf/). This system assimilates observations four times daily from radiosondes, aircraft data (ACARS), cloud drift winds, ASOS stations, buoys and ships. Furthermore, after the 00 and 12 UTC assimilation cycles, all 90 ensemble members are integrated for 24 hours. Along with probabilistic information about the forecast, this dataset also provides estimates of the sensitivity of a given forecast field to the analysis and future observations. Verification statistics for this real-time system will be discussed and compared to the GFS.